Benchmarking Technical Properties and Aesthetic Reproduction of Five Major Image-Generative AI Applications: A Comparative Study of Google Gemini, ChatGPT, Meta AI, Microsoft Copilot, and Mistral AI

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

AbstractThis study evaluates the output characteristics of five prominent image-generative Artificial Intelligence (AI) applications—Google Gemini, ChatGPT, Meta AI, Microsoft Copilot, and Mistral AI—to establish objective benchmarks for end-users. Using a standardized prompt across all platforms, the research analyzed both qualitative aesthetic reproduction of four distinct artistic techniques (Anime, Tenebrism, Decorative Flatness, and Sfumato) and quantitative technical metadata. Findings reveal that Meta AI provides the highest resolution (2.36 million pixels) while maintaining high efficiency through the WebP format. In contrast, Google Gemini prioritizes professional post-processing through 32-bit PNG outputs. Mistral AI exhibited a model-level bias toward landscape aspect ratios regardless of instructions. These results indicate that AI applications are strategically differentiated by their underlying technical design philosophies, ranging from web-native efficiency to professional-grade material production. Researcher’s Position and Conflict of InterestThis study is categorized as Citizen Science. The author is an independent researcher with no financial or organizational affiliations with Google LLC, OpenAI, Meta Platforms, Inc., Microsoft Corporation, or Mistral AI SAS. There are no conflicts of interest to declare.

Links and sources

Comments